Content About Stocktwits
No Image Available

‘Gaming the System’ for AI. Nvidia, OpenAI & Anthropic. ARD #138

Author
Michael Parekh AI: Reset to Zero
21 hours ago
11 MIN READ

Today’s theme is ‘Gaming the System’ for AI, and I mean it in both senses. The skeptics use the term as an accusation: circular financing, manufactured demand, risk quietly moved onto other people’s books. All real financial engineering concerns, seen in every tech wave, not just the modern run from PCs to internet to mobile to AI, but the waves hundreds of years ago too: railroads, industrialization, electricity.

I mostly mean it the other way. The best operators through every cycle make sure their businesses and strategies are accommodated to the capital markets, private and public: the mechanics of funding, debt, equity, and every constituent that can help in the financing. That is what is going on in spades in today’s AI Tech Wave.

Three events today: Nvidia, OpenAI and Anthropic. My take on each, then my Overall Take. Plus a Gadget AI on Apple’s glassy 20th anniversary iPhone, and your questions.

(1) Nvidia Assembles a ‘Who’s Who’ Finance Consortium for AI Infrastructure

MP TAKE: Four days ago I wrote about Nvidia the ‘Kingmaker’, now ‘uber AI investor and banker’. That was Nvidia lending its own ballooning balance sheet. Today’s move is the next turn of the dial: mobilizing OTHER people’s money, half a trillion dollars of AAA institutional capital pointed at AI factories. When the funding system you have runs out of road, you build new roads. That is the game, played at the highest level.

The mechanics, per the FT: Nvidia is partnering with the who’s who of world finance on a funding package of more than $500 billion for AI infrastructure over the coming years. The names are the top of the institutional list: Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs (my alma mater) and KKR, among others. Six memorandums of understanding, dedicated pools of capital, financing Nvidia’s ecosystems at ‘attractive rates for Nvidia customers’, in their words. Nvidia calls it the first agreement of its kind, and having tracked this stuff for decades, I would agree.

Jensen’s framing, in his own words:

“We began by building chips; today, we are helping create a new class of productive, investable infrastructure: AI factories.”

The context from my Kingmaker piece: $750 billion of announced deals, and a $250 billion backstop under discussion so OpenAI can lease 10 gigawatts of SoftBank-built data centers in Ohio ahead of its IPO. Nvidia writing checks that grow with its market share. The new consortium points the biggest pools of long-term capital, insurance, pensions, retail infrastructure funds, at the same build-out.

Why now? Because the old funding system is maxing out. The hyperscalers blew through some of the best free cash flows on the planet. Google went to equity, $85 billion of it, with Berkshire in. Morgan Stanley sees $3.5 trillion of hyperscaler spend through 2028; Apollo’s president calls the build-out more than $8 trillion of capital. The echoes, as Ben Thompson’s Stratechery piece lays out, are the railroad financings of the 1870s, about equal in today’s dollars, with all their ups and downs. Those risks deserve the scrutiny they are getting.

And do not miss the demand side of the same game, per The Information: Nvidia is pouring resources into Nemotron, its open source AI models push, aiming for the best open models in the world. Financing the supply, seeding the demand. Nvidia is not just playing the AI game. It is redesigning the board, and adding a few squares to it.

Sources:

For longtime readers:

(2) OpenAI Prunes Shareholders Ahead of its Mega-AI IPO

MP TAKE: This is classic pre-IPO engineering, and I say that fondly: as a former internet analyst at Goldman Sachs, I am very familiar with the genre. Prune the shareholder list before the roadshow so the register holds strong hands, holders who will not be chomping at the bit to sell at the first lockup. There are just more zeros in this tech wave’s version.

The mechanics, per Bloomberg: OpenAI bought back about $7 billion of shares from current and former employees in a tender offer, and notably bought them back itself, from its $122 billion-plus of prior raises, rather than lining up outside investors as in past tenders. The deal valued the company at $852 billion, unchanged from its most recent round, versus about $965 billion for Anthropic. OpenAI confidentially filed for its IPO in June.

The race context: Anthropic is likely out of the gate sooner, and OpenAI is expected between late this year and next. OpenAI has a lot of wood to chop against Anthropic’s Claude Code and Cowork in AI Coding and enterprise, where OpenAI says about half its business now sits with its Codex offerings. They threw in the ‘Kitchen Sink’ on costs weeks ago, one of the things you do prepping for an IPO. This is the next step on that road.

Sources:

For longtime readers:

(3) Anthropic Ramps Compute as it Readies its Mega-AI IPO

MP TAKE: Anthropic has the high-class problem of a hit product, Claude Code and Cowork, that eats huge amounts of compute, with less compute laid on than OpenAI. That is OpenAI’s tactical advantage, and Anthropic is closing the gap deal by eye-popping deal. These moves are operational preparation for the IPO: showing investors the supply is locked for the demand. The game Anthropic absolutely has to play to keep its lead.

The mechanics, per Bloomberg: a $9.1 billion, 20-year compute deal with Riot Platforms, a Bitcoin mining company turned AI data center supplier, for 191 megawatts from its Rockdale, Texas campus, running through June 2048 with extensions that could take it to $16.1 billion. Riot’s shares jumped 25% on the news. Add roughly $45 billion of computing from Elon’s companies at premium prices, a $10 billion deal with months-old Volta Infra, and the SpaceX and Google compute deals from the WSJ’s pre-IPO piece.

And the suppliers are part of the financing game too: Nvidia and Google provide backstops and financial capabilities along the way, the same pattern as the consortium above. Everyone in this wave runs both sides of the deal table.

Sources:

For longtime readers:

MP OVERALL TAKE: Step back and notice: none of today’s three events is about shipping models. All three are about arranging capital and ramping infrastructure for anticipated demand, with multi-year supply chains for chips, memory and power, at roughly $50 billion per gigawatt of AI capacity.

The echoes of the railroad build-outs of the late 1800s are real: those took two to three decades, with huge ups and downs, and the risk had to be managed against the rewards across balance sheets and pools of capital. Same here. We covered the supply side in part 1 and the demand side in part 2 of our compute series this week. Today was the financing side, ramping in earnest into Anthropic’s IPO potentially this Fall, and OpenAI’s after.

Gaming the system, in the best sense. Until the system says otherwise.

GADGET AI

A More Glassy 20th Anniversary iPhone Lineup in 2027

MP TAKE: Next year’s iPhones matter for the 20th anniversary of Steve Jobs’ high conviction bet on a radically new idea for a smartphone. Back then it was also, ironically, about glass: a full glass touch screen no one had seen, when we were all BlackBerry-holics, and mainstream users had to learn ‘Pinch and Zoom’ by the billions. But the core innovation I am focused on is Apple Intelligence with Siri AI: personal AI in the cloud and on the phones, training on our personal data in a secure, safe, trusted way, relative to what Google or Meta can do. That is the new glassy surface Apple can truly slide to consumer AI success.

The story, per Bloomberg’s Mark Gurman: Apple’s glass-centric 20th anniversary iPhone remains on track for 2027, with plenty of design-leak detail for smartphone-aholics like yours truly.

Sources:

For longtime readers:

Q1: What’s the one iPhone that brings daily joy to MP?

My daily driver is the iPhone 17 Pro Max in orange. But my secondary phone, the one I find every excuse to use, is the iPhone Air. Very light, a slightly lesser camera, lower battery life, and none of that matters: interacting with a device that thin and light feels fundamentally different, and you notice its delightful qualities every day. That feeling is about to repeat across this fall’s smartphone cycle.

Q2: What’s the closest equivalent for MP on the Android side?

The original Google Pixel Fold with its passport design, still delightful over three years later, opening to a 7.6 inch screen, about the same as Samsung’s new wide Galaxy Z Fold 8. With ‘RAMageddon’ pushing new folds to $2,000 and up, you can pick up the original Pixel Fold for about $500 and get most of the experience. Watch secondhand phones become a real trend over the next two or three years. And I am waiting eagerly for the wide Apple Fold phones.

WRAP

Today’s AI-RTZ #1175 is on ‘RAMageddon’ lobbying Washington DC, and Apple vying for China’s CXMT, directly relevant to today’s gadget talk: the memory crunch hits devices and clouds alike. And tonight’s AI-RTZ #1176 goes deeper on Anthropic warming up investors for its mega-AI IPO.

These are the impacts of the AI Tech Wave we track every day for you, the AI Curious Folk. Stay tuned.

Tomorrow, it’s ARD 139 and AI-RTZ #1176.

Thanks for joining us today, AI Curious Folk.

— MP

Full Source Reading

For those who want to go deeper on today’s stories:

(1) Nvidia’s finance consortium

(2) OpenAI’s shareholder pruning

(3) Anthropic’s compute ramp

Gadget AI: the glassy 20th anniversary iPhone

Clips from today

Clip 1. Apple’s Folding Phone Strategy

Apple rumored to go passport-wide with its first Fold, Samsung’s wide Galaxy Z Fold 8 drawing raves, and my three-year-old Pixel Fold still delightful.

MP Take: with ‘RAMageddon’ pushing new folds past $2,000, watch secondhand phones become the trend of the next two or three years.

Clip 2. Nvidia’s $500B Wall Street AI Consortium

The who’s who of world finance: Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR. Six MOUs, dedicated pools of capital, at attractive rates for Nvidia customers.

MP Take: the first agreement of its kind. AI factories, financed by Wall Street’s biggest.

Clip 3. Nvidia, Now the AI Investor & Banker

From the $750 billion Kingmaker deals and the $250 billion OpenAI backstop to mobilizing half a trillion dollars of other people’s money.

MP Take: when the funding system maxes out, the best operators build a new system. That is the game.

Clip 4. iPhone Air: My Joyful Secondary Phone

The phone I find every excuse to use: very light, and it feels fundamentally different in the hand every day.

MP Take: that feeling repeats across this fall’s smartphone cycle as the folding phones arrive in force. The Air is the preview.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)





Want the latest?

Sign up for Michael Parekh's Newsletter below:


Subscribe Here

More like this

BROWSE ALL ARTICLES